About Truespar
Infrastructure for AI builders.
Who We Are
Truespar is a research arm of The Intelligence Company AB (publ), a Swedish company providing decision intelligence platforms to enterprises. The research question is simple: how much performance can modern hardware actually give? We answer it by building infrastructure engines from scratch and measuring them against the software they replace. The engines that prove out become products and run on our own production infrastructure - that is where Paddock, Traverse, and Torque come from. Research is how we build; what ships is a supported product that has already carried real workloads in our production.
What We Build
We build the layers an AI product runs on: model inference (Paddock), databases (Traverse), and search (Torque), with caching (Forge) and file storage (Trove). Every engine runs on your own machines and installs in minutes: download one file and start it. Each one exposes an API compatible with the tool it replaces - OpenAI and Anthropic, Neo4j, Typesense - so switching is a config change, not a rewrite. It also keeps your data and your costs inside your own network, which is what enterprises now call sovereign AI.
Why We Build It
An AI product is only as fast and as cheap as the infrastructure under it. Agents hit that infrastructure harder than people ever did: one agent can generate more queries, searches, and model calls in a minute than a human user produces in a day. Most enterprise software was architected before modern hardware existed - hundreds of cores, fast memory, GPUs - so it leaves performance on the table and charges you for the privilege. The usual answer is to pay a cloud provider to absorb it, and that bill grows with every agent you add. Ours is faster software: engines that use everything the machine has, with SIMD code paths matched to the CPU's instruction set and hand-written GPU kernels tuned per NVIDIA chip generation.
The Name
Truespar is two words. True is the accuracy and reliability of our platforms. Spar is the challenge: proving them under pressure, in the open, against the tools they replace.
What We Believe
Eliminate Every Layer
Every unnecessary abstraction is latency. We strip away what doesn't earn its cost and let software talk directly to hardware.
We Run It in Production
Our engines run in our own production first. They carry real load and meet enterprise SLAs before anyone else depends on them.
Built to Our Own Standards
We build to our own bar for uptime, speed, and ease of operation. What we learn in production flows back into the engines.